CS B553: A O L

unknownlippsΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 3 χρόνια και 11 μήνες)

60 εμφανίσεις

CS B553: A
LGORITHMS

FOR

O
PTIMIZATION

AND

L
EARNING

aka “Neural and Genetic Approaches to Artificial
Intelligence”

Spring 2011

Kris Hauser

T
ODAY

S

A
GENDA


Topics covered


Prerequisites


Class organization & policies


Coursework


Math review

W
HAT

IS

O
PTIMIZATION
?


The problem of choosing the “best” solution from
some set of candidate solutions


Airplane wing that minimizes drag


Stock portfolio that maximizes return on investment


Feedback control strategy with highest probability of
picking up an object


(In many problems, it is easier to measure the quality
of candidate solutions than to produce the optimum!)


A mathematical
discipline that is
heavily
studied
and utilized in
other
fields


Powerful idea in AI, machine learning, computer
vision, engineering, economics, applied sciences

O
PTIMIZATION

L
EARNING

O
BJECTIVES


Hands
-
on experience in specifying
mathematical
optimization problems


Defining objective functions, constraints


Identifying problem characteristics (e.g., convexity)


Characteristics of small/medium/large scale problems


Mostly continuous optimization, some discrete and
mixed
-
integer optimization


Solving optimization problems in practice


Algorithms: descent
-
based, simplex based, stochastic


Software packages


Performance tricks


Applied to realistic scenarios



W
HAT

IS

L
EARNING
?


Deriving “meaningful” quantities from raw data (e.g.,
gathered from logs,
surveys,
sensors) and employing
them


Diagnosing a patient from reported symptoms


Recognizing human activity from video


Forecasting weather or economic behavior from history


Diverse range of learning tasks, most of which involve
one or more of:


Fitting a model by adjusting model parameters


Selecting a model structure that explains the data


Using a model to infer meaningful quantities


Many learning tasks are essentially optimization
problems!

L
EARNING

L
EARNING

O
BJECTIVES


Conceptual frameworks for large scale learning


Graphical
models (e.g., Bayesian networks)


Hidden Markov
Models (HMMs),


Dynamic Bayesian Networks (DBNs)


Understanding of key components for
implementing many learning algorithms


Belief propagation


Expectation maximization algorithms


Monte Carlo techniques


Experience applying algorithms to real
-
world
datasets

O
RGANIZATION


http://www.cs.indiana.edu/classes/b553
-
hauserk



Lectures, readings


Lecture notes for optimization unit


Probabilistic Graphical Models: Principles and
Techniques
(
Koller

and Friedman)

for learning unit


In
-
class group exercises

C
OURSE

WORK


Attendance and participation: 20% of grade


8
homework assignments (4 written, 4
programming
):
80% of grade


Programming in language of your choosing


Optional final
project


Original research, survey, or reproduction of recent
research paper with a substantial
optimization/learning component


Counts
for 4 HW grades


P
OLICIES


Office hours: W 10
-
11am, Info E 257 (or by
appointment)


Should respond to email in 24 hours


Late HW:


10% deducted for every day late

P
REREQUISITES


CS B551 or equivalent introduction to AI course.
Specifically,
probabilistic reasoning
and Bayesian
networks


Calculus. (Multivariate recommended)


Linear algebra


L
ECTURE